Three different tests

Recognition means the system describes a brand correctly when its name is already present. Discovery means the brand appears without a hint in a relevant competitive set. Recommendation adds an argument by connecting the offer to the buyer's criteria, risks or constraints.

Why one prompt proves nothing

The answer can change with wording, language, region, active search, product version and generation variance. A defensible result needs repeats, paraphrases of the same intent, control competitors and saved answers with a timestamp.

Knowledge is not commercial presence

A system may accurately summarise an About page or LinkedIn profile and still omit the brand from a supplier-selection question. The second task needs a clear category, evidence of fit, independent mentions and sources that enter the search or retrieval context.

What we measure instead of a decorative score

We record appearance across an agreed scenario set, the role assigned to the brand and the competitor or source gap. Each observation can be checked again. A single composite score without raw answers hides where the failure occurred.

The evidence boundary

GEO research shows that the form and evidence of a document already present in context can alter how it is used. That does not prove guaranteed organic discovery. Operationally, access, retrieval, mention, citation and commercial outcome must remain separate stages.

Practical takeaway

Build a three-level matrix: recognised, discovered and meaningfully recommended. Save the answers, repeats, sources and competitors for every level. Then content or external-source work has a defined objective instead of a vague goal to increase an AI score.

Primary sources

These links point to platform documentation and research supporting the claims in this article.

  1. GEO: Generative Engine Optimization research
  2. Critical survey of GEO research, 2023-2026
  3. Google: how AI features find supporting pages